We have deep expertise in machine learning, audio, and speech processing and are using that to create innovative products to help people hear better.
Our first product, HeardThat, turns your smartphone into a sophisticated hearing assistive device that brings out the conversation in noisy social situations.
HeardThat is a subsidiary of Singular Software, and is located in the beautiful city of Vancouver, British Columbia, Canada. Singular Software was a Top 10 finisher in the 2019 New Ventures BC competition.
The Noise Problem
Noise is everywhere and more so today than ever before. It makes it hard for anyone to hear conversation in coffee shops, restaurants, at business functions, etc. For people with hearing loss, it can be an impossible situation.
Traditional hearing assistive products don’t solve the problem. They don’t have the power or ability to know what is speech and what is noise, are generally only able to filter and amplify sounds.
A Better Solution
HeardThat takes advantage of machine learning to bring a new kind of solution for noise. Traditional noise cancellation doesn’t know what noise is, it just reduces sounds of all kinds, including speech. The sophisticated algorithms that make HeardThat work separate speech from noise, get rid of the noise, and just deliver the speech to you. This is powerful stuff that can’t be run on small wearable devices.
Instead, HeardThat works with the devices you already have: your smartphone and your favorite listening devices, including hearing aids. Just start the app, lay your phone on the table, and listen to the audio coming from the phone to your ears.
The Science
Traditional speech enhancement uses a combination of characterizing noise, modelling speech, and applying frequency-based suppression and enhancement. These are very hard to do.
Machine learning, and in particular, deep learning, takes a different approach. Instead of trying to mathematically model sound, neural networks are trained by giving them thousands of hours of examples of clean speech and noisy speech. The models learn what is speech and what is noise, and the result is a significantly new way to separate the two.